Commit History
-
[KM-691] Traceability: add resolved data_used layer for the FE
Rifqi HafizuddinGET /api/v1/traceability now carries a user-facing data_used[] block so the FE renders real source/table/column names instead of raw IR ids. One entry per retrieve_data call, resolved deterministically from the catalog (no LLM, never-throw): - data_used: source / tables / joins / columns_read / output_columns / filters / group_by / order_by / query. Result columns split into read-from-data (kind=column) vs computed (kind=computed + formula, no id), so an aggregate alias like total_revenue is never shown as a real column. - ids are kept but machine-only (documented; FE must not render them). - tool_calls[].summary: plain-English per-step headline (fixed template, no LLM). - sources[]: add source_name + all tables touched (incl. join targets). - new src/traceability/resolve.py; catalog threaded to the scratchpad at the slow-path composition root; order_by alias + unresolved-id fallbacks handled.a646735